Download
D4RL (Datasets for Deep Data-Driven Reinforcement Learning) is a benchmark suite focused on offline reinforcement learning — i.e., learning policies from fixed datasets rather than via online interaction with the environment. It contains standardized environments, tasks and datasets (observations, actions, rewards, terminals) aimed at enabling reproducible research in offline RL. Researchers can load a dataset for a given task (e.g., maze navigation, manipulation) and apply their algorithm without the need to collect fresh transitions, which accelerates experimentation and comparison. The API is based on Gymnasium (via gym.make) and each environment also exposes a method get_dataset() that returns the offline data to learn from. The repository emphasizes open science, reproducibility, and benchmarking at scale, making it easier to compare algorithms on equal footing.
Features
- Offline reinforcement-learning benchmark suite with fixed datasets and tasks
- gym.make compatible environments plus _get_dataset() method returning transitions
- Support for algorithm comparison, reproducibility and standardized tasks
- Large variety of tasks (navigation, manipulation, maze, robotics) and datasets
- Open licensing (Apache-2.0 for code, CC BY for data) to facilitate research and attribution
- Used widely in RL research for benchmarking offline learning algorithms
Project Samples
